Fig. 1 Two strong signals. 10.681 MHz and 10.716 MHz with a Delta 44 soundcard (unmodified) under Microsoft Windows with the sampling speed set to 96 kHz. 
Fig. 2 Two strong signals. 10.681 MHz and 10.716 MHz with a Delta 44 soundcard (unmodified) under Microsoft Windows with the sampling speed set to 100 kHz. |
|
From these figures it is obvious that the Windows sound driver
uses the Delta 44 at 48 kHz because the 10.681 MHz signal is missing.
The software upsampling to 99.373 kHz increases the noise floor
and adds spurious signals.
In Windows one can choose the compromise between resampling quality
and resouces utilized for that.
The control is at |Control Panel|sound and Audio Devices|Advanced|.
This control is available in some Windows installations depending
on update status and presumably drive routines installed.
Unfortunately this control does not affect the mediocre quality
of the (nominal) 100 kHz signal from a Delta 44 even in
cases when the control is present (at least not under Win 2000.)
If you have a high performance system with a Delta44 or similar soundcard for high speed input you need a second soundcard for the output. The Delta 44 will not allow different sampling speeds for input and output and Linrad is not designed for high speed output. The A/D sampling rate has a large influence on the work load for the CPU. On modern computers (Pentium III and up) CPU power is usually no issue so it does not matter if you sample faster than necessary. To process normal SSB bandwidth a sampling rate in the range 5000 to 8000Hz is sufficient. Heavy oversampling at low bandwidth, like 44100Hz for an SSB receiver may improve the dynamic range of the digital processing but that is of no value at all because an SSB receiver is limiting the dynamic range within the passband anyway. Already without oversampling the computer has a much better dynamic range than an SSB receiver. If your analog hardware provides a real signal you should start by selecting the lowest possible pitch. Place the BFO as close as possible to the passband, not further away than necessary to get sufficient attenuation of the mirror image (the other sideband). This way you make the highest frequency as low as possible. As an example, select 6000Hz for the sampling speed. The Nyqvist frequency is then 3000Hz, and that is the highest frequency that can be present in the digital signal. Assuming the analog hardware is a normal SSB radio, the signal spectrum is flat between 200 and 2600Hz or so. At 3000Hz the level has dropped by perhaps 20dB and at 3300Hz the signal is down by perhaps 50dB. A signal at 3300Hz will show up as a signal at 2700Hz in the digital representation but the anti alias filter inside the soundcard has suppressed it further so it is well below -50dB when it shows up as an alias spur. In case the mirror image gives spurs, move the BFO by 1 kHz in this example and set the sampling speed to 8000 Hz. Performance at the high side will be the same but the mirror image between 0 and 2.2 kHz will disappear. Make a preliminary selection based on what you think about your analog hardware. When the setup is finished, Linrad is an excellent spectrum analyser and you can easily evaluate the spur levels and decide if you can reduce the sampling rate or possibly gain dynamic range by increasing it. It will also be easy to set the BFO pitch for good suppression of the image spur with minimum offset from the passband. If your A/D converter has more than 16 bits you will be prompted for A/D data format. You may choose 16 or 24 bits. Even if you have a 24 bit soundcard like Delta44 it is not obvious what choice to make. The Delta44 soundcard is said to be a 24bit A/D converter but that is not true at all. When run at 96kHz it is perhaps 16.5 bits and the loss in dynamic range by selecting 16 bit format is very small. Selecting 32 bits causes a minimal increase in processor load, typically less than 1%. The disadvantage of using 32 bit format is that saving data on the hard disk is somewhat slower. In 32 bit mode Linrad saves raw data in a 18bit format which uses 12.5% more disk space than the 16 bit format. As it happens, my PentiumIII can save data for long periods of time in 16 bit format, but with 12.5% more disk load, the disk writes take too much time and A/D overrun starts to become a problem. When Linrad knows what device to use for input and how to open it, the device is tested. If all is OK the next parameter to set is the radio interface. the choices are: 1: One rx channel, normal audio. 2: One rx channel, I/Q stereo (direct conversion rx) 3: Two rx channels (adaptive polarisation/phasing) 4: Two rx channels, I/Q stereo (adaptive polarisation/phasing) The fourth line is present only if your input device has 4 A/D channels. (Or if you have opened two stereo channels on the same soundcard) Normal audio means a real valued signal such as the audio output of a normal radio receiver. I/Q stereo means a complex signal that uses two A/D channels to send the in phase signal and the quadrature signal from a direct conversion radio to the soundcard. Select the mode that fits your analog hardware. Finally you have to select an output device driver. The output sampling speed and number of bits is not set here among the general parameters since different processing modes may need different formats. As an example wideband AM may need 24 kHz to provide good sound quality on music while morse code does not even need 5000 Hz, the lowest speed I have seen supported by hardware. Another example: When saving the loudspeaker output from Linrad there is no need to use more than 8 bits if S/N is very low. That would make the saved file twice smaller. Do not forget to press W to save your new parameter selection in the dsp_uiparm file. SM 5 BSZ Home page Linrad home page |